tms470-compiler:Linker warning using GROUP

Please show me the lines in your link command file which set up that GROUP, as well as the lines in the map file which show how the linker allocated it.  You are building for an ARM device, right?  Are you building for EABI?  

Thanks and regards,

-George

Hi

Georgem said:

Please show me the lines in your link command file which set up that GROUP, as well as the lines in the map file which show how the linker allocated it.  You are building for an ARM device, right?  Are you building for EABI?  

Thanks and regards,

-George

It was all in the first post, but the forum-software is crap and wants me to insert newline in HTML code :(
It is ARM and eabi, yes.

Anyway
GROUP:
{
/* initialized data */
.data
.bss
} load = rom, run = system_mod,RUN_START(system_mod_start),RUN_SIZE(system_mod_initsize)

And the map file:
.data 0 c0000000 00000018
c0000000 00000014 sconf.obj (.data)
c0000014 00000004 hello.obj (.data)

.bss 0 c0000018 00000c80
c0000018 00000c2c sciopta.obj (.bss)
c0000c44 00000040 resethook.obj (.bss) [fill = 0]
c0000c84 00000014 sconf.obj (.bss) [fill = 0]

LINKER GENERATED COPY TABLES

__TI_cinit_table @ 80004b4c records: 3, size/record: 8, table size: 24
.data: load addr=80004b08, load size=00000019 bytes, run addr=c0000000, run size=00000018 bytes, compression=rle
.bss: load addr=80004b28, load size=0000000b bytes, run addr=c0000018, run size=00000c80 bytes, compression=rle

Moving the load...,run.. direct after GROUP : does not change a thing.

42Bastian said:

It was all in the first post, but the forum-software is crap and wants me to insert newline in HTML code

Sorry about that.  Next time, just put it in a text file and attach it to the post.  The file attachment buttons can be seen by clicking on the "Options" link near the upper left corner of the post composition window.

Well, .data is an initialized section, and .bss is an uninitialized section.  When you put them together in the same GROUP, things can get weird.  Just to see if it makes things better, please move the .bss section outside the GROUP.  Let me know what happens.

Thanks and regards,

-George

 

 

Hi George,
I tried it, but the warning stays. Worse, my RUN_SIZE() label does no longer include the size of the BSS. Regards,

I now suspect a bug in the linker.  Unfortunately, this means we need a test case we can build and see the error for ourselves.  That's often challenging for linker issues.  I'd appreciate if you would put a test case together.  Please include the exact tools version and build options.

Thanks and regards,

-George

 

Hi.
ok. I will try to set up a test case ASAP.

Hi George, attached a simple project build with File->New->CCS Project. I choose a Stellaris linker command file and just added GROUP and load info. Added simple test.c with data and bss section.
I use 4.6.4 tools for ARM.
CCS is 4.1.3.00038
BTW: The forum interface does not allow 7zip.

Thank you for the test case. I submitted SDSCM00038199 in our SDOWP system. Feel free to track it at the SDOWP link in my sig below.

A workaround is to rewrite the GROUP like this ...

    GROUP : {
       .data   load = FLASH
       GROUP {
          .bss 
          .sysmem 
          .stack  
       }
    } run = SRAM

I think that does what you want.

Thanks and regards,

-George

Hi George,
the WA does not work. Though it places all section one after another and RUN_SIZE(),RUN_START() are correct, the linker does not generate init-data for the .data section (maybe another bug).
There are other issues :(
If I change the line

  .data load = FLASH,table(BINIT, compression = rle)

and add

 .binit > FLASH

I get init-data for the .data section but if I GROUP the flash stuff:

 GROUP : {
    .text
    .const
    .cinit
    .pinit
    .binit} > FLASH

I get this:
warning: output section ".data" refers to load symbol "_nop" and hence cannot be compressed; compression "rle" is ignored
Note: Without GROUPing, the .data section was compressed.
If .cinit and .binit are not part of the GROUP, .data is compressed.
BTW: Checking for size == 0 to decide whether a section is compressed or not is bad habit, as with table() it is possible, that zero-length sections are place in the table.

  GROUP : {
    .data  load = FLASH 
    GROUP: {
    .bss 
    .sysmem 
    .stack  
    }
    } run = SRAM, RUN_SIZE(rs),RUN_START(rstart),table(mytab),table(BINIT,compression = rle)

==> see sysmem

binit @ 000004d0 records: 4, size/record: 12, table size: 52
	.data: load addr=00000508, load size=00000018, run addr=20000000, run size=00000018, compression=off
	.bss: load addr=20000018, load size=00000190, run addr=20000018, run size=00000190, compression=none
	.sysmem: load addr=200001a8, load size=00000000, run addr=200001a8, run size=00000000, compression=none
	.stack: load addr=200001a8, load size=00000100, run addr=200001a8, run size=00000100, compression=none
mytab @ 00000448 records: 4, size/record: 12, table size: 52
	.data: load addr=00000430, load size=00000018, run addr=20000000, run size=00000018, compression=none
	.bss: load addr=20000018, load size=00000190, run addr=20000018, run size=00000190, compression=none
	.sysmem: load addr=200001a8, load size=00000000, run addr=200001a8, run size=00000000, compression=none
	.stack: load addr=200001a8, load size=00000100, run addr=200001a8, run size=00000100, compression=none

Oh, there is more stuff not working:

.data  load = FLASH ,table(BINIT, compression = rle),table(mytab, compression = rle)

Gives a warning and disables compression. Regards,

I tried George's WA and was able to produce an executable with initialized data for the .data section using the 4.6.4 linker.   Can you attach your .out file?  I'm not sure why your .data section is empty.  We need to examine the output file to see exactly what's going on with any case where you get unexpected output.

Regarding your original post, the reason why the warning is emitted from the linker is because you should not have a GROUP that specifies both  the load and run placement when that GROUP has an uninitialized section member.

-Mack

The exact linker warning you got in your original post is a valid warning.  Because you are using EABI and are auto initializing variables at runtime (linker option -c enabled by default), the linker emits this warning since there are no sections in the GROUP that have load data.  You can avoid this warning (if you actually want .data to have a different load and run address) for the 4.6.4 release by specifying that the .data section should not be autoinitialized:

GROUP: {.data: type=NOINIT .bss } load = rom, run = system_mod

-Mack